Blackfin optimized YUV420 to RGB CSC Color Space Converters.

YUV2 -> RGB BGR for 565, 555 and 888 a.k.a. 24bit color.
Speed-up compared to C version compiled with -O3  187.28%
Patch by Marc Hoffman %mmh A pleasantst P com%
Original thread:
Date: May 9, 2007 2:46 AM
Subject: [FFmpeg-devel] PATCH BlackFin yuv2rgb color space conversion

Originally committed as revision 23307 to svn://svn.mplayerhq.hu/mplayer/trunk/libswscale

libswscale/Makefile

@@ -12,6 +12,9 @@ OBJS= swscale.o rgb2rgb.o
 OBJS-$(TARGET_ALTIVEC)     +=  yuv2rgb_altivec.o
 OBJS-$(CONFIG_GPL)         +=  yuv2rgb.o
 
+OBJS-$(TARGET_ARCH_BFIN)     +=  yuv2rgb_bfin.o
+ASM_OBJS-$(TARGET_ARCH_BFIN) += internal_bfin.o
+
 HEADERS = swscale.h rgb2rgb.h
 
 include ../common.mak

libswscale/swscale.c

@@ -1992,7 +1992,7 @@ SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH
 #endif
 
 #if !defined(RUNTIME_CPUDETECT) || !defined (CONFIG_GPL) //ensure that the flags match the compiled variant if cpudetect is off
-    flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC);
+    flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
 #ifdef HAVE_MMX2
     flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
 #elif defined (HAVE_3DNOW)
@@ -2001,6 +2001,8 @@ SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, int dstW, int dstH
     flags |= SWS_CPU_CAPS_MMX;
 #elif defined (HAVE_ALTIVEC)
     flags |= SWS_CPU_CAPS_ALTIVEC;
+#elif defined (ARCH_BFIN)
+    flags |= SWS_CPU_CAPS_BFIN;
 #endif
 #endif /* RUNTIME_CPUDETECT */
     if (clip_table[512] != 255) globalInit();

libswscale/swscale.h

@@ -74,6 +74,7 @@ extern "C" {
 #define SWS_CPU_CAPS_MMX2     0x20000000
 #define SWS_CPU_CAPS_3DNOW    0x40000000
 #define SWS_CPU_CAPS_ALTIVEC  0x10000000
+#define SWS_CPU_CAPS_BFIN     0x01000000
 
 #define SWS_MAX_REDUCE_CUTOFF 0.002
 

libswscale/swscale_internal.h

@@ -162,6 +162,22 @@ typedef struct SwsContext{
 
 #endif
 
+
+#ifdef ARCH_BFIN
+    uint32_t oy           __attribute__((aligned(4)));
+    uint32_t oc           __attribute__((aligned(4)));
+    uint32_t zero         __attribute__((aligned(4)));
+    uint32_t cy           __attribute__((aligned(4)));
+    uint32_t crv          __attribute__((aligned(4)));
+    uint32_t rmask        __attribute__((aligned(4)));
+    uint32_t cbu          __attribute__((aligned(4)));
+    uint32_t bmask        __attribute__((aligned(4)));
+    uint32_t cgu          __attribute__((aligned(4)));
+    uint32_t cgv          __attribute__((aligned(4)));
+    uint32_t gmask        __attribute__((aligned(4)));
+#endif
+
+
 } SwsContext;
 //FIXME check init (where 0)
 

libswscale/yuv2rgb.c

@@ -611,6 +611,14 @@ SwsFunc yuv2rgb_get_func_ptr (SwsContext *c)
     }
 #endif
 
+#ifdef ARCH_BFIN
+    if (c->flags & SWS_CPU_CAPS_BFIN)
+    {
+        SwsFunc t = ff_bfin_yuv2rgb_get_func_ptr (c);
+        if (t) return t;
+    }
+#endif
+
     av_log(c, AV_LOG_WARNING, "No accelerated colorspace conversion found\n");
 
     switch(c->dstFormat){

libswscale/yuv2rgb_bfin.c

+/*
+ * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
+ *                    April 20, 2007
+ *
+ * Blackfin Video Color Space Converters Operations
+ *  convert I420 YV12 to RGB in various formats,
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+#include <assert.h>
+#include "config.h"
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
+#include <unistd.h>
+#include <bits/bfin_sram.h>
+#include "rgb2rgb.h"
+#include "swscale.h"
+#include "swscale_internal.h"
+
+
+#define L1CODE __attribute__ ((l1_text))
+
+extern void ff_bfin_yuv2rgb555_line (uint8_t *Y, uint8_t *U, uint8_t *V, uint8_t *out,
+                                     int w, uint32_t *coeffs) L1CODE;
+
+extern void ff_bfin_yuv2rgb565_line (uint8_t *Y, uint8_t *U, uint8_t *V, uint8_t *out,
+                                     int w, uint32_t *coeffs) L1CODE;
+
+extern void ff_bfin_yuv2rgb24_line (uint8_t *Y, uint8_t *U, uint8_t *V, uint8_t *out,
+                                    int w, uint32_t *coeffs) L1CODE;
+
+typedef void (* ltransform_t)(uint8_t *Y, uint8_t *U, uint8_t *V, uint8_t *out,
+                              int w, uint32_t *coeffs);
+
+
+static void bfin_prepare_coefficients (SwsContext *c, int rgb, int masks)
+{
+    int oy;
+    oy      = c->yOffset&0xffff;
+    oy      = oy >> 3; // keep everything U8.0 for offset calculation
+
+    c->oc   = 128*0x01010101U;
+    c->oy   =  oy*0x01010101U;
+
+    /* copy 64bit vector coeffs down to 32bit vector coeffs */
+    c->cy  = c->yCoeff;
+    c->zero = 0;
+
+    if (rgb) {
+        c->crv = c->vrCoeff;
+        c->cbu = c->ubCoeff;
+        c->cgu = c->ugCoeff;
+        c->cgv = c->vgCoeff;
+    } else {
+        c->crv = c->ubCoeff;
+        c->cbu = c->vrCoeff;
+        c->cgu = c->vgCoeff;
+        c->cgv = c->ugCoeff;
+    }
+
+
+    if (masks == 555) {
+        c->rmask = 0x001f * 0x00010001U;
+        c->gmask = 0x03e0 * 0x00010001U;
+        c->bmask = 0x7c00 * 0x00010001U;
+    } else if (masks == 565) {
+        c->rmask = 0x001f * 0x00010001U;
+        c->gmask = 0x07e0 * 0x00010001U;
+        c->bmask = 0xf800 * 0x00010001U;
+    }
+}
+
+static int core_yuv420_rgb (SwsContext *c,
+                            uint8_t **in, int *instrides,
+                            int srcSliceY, int srcSliceH,
+                            uint8_t **oplanes, int *outstrides,
+                            ltransform_t lcscf, int rgb, int masks)
+{
+    uint8_t *py,*pu,*pv,*op;
+    int w  = instrides[0];
+    int h2 = srcSliceH>>1;
+    int i;
+
+    bfin_prepare_coefficients (c, rgb, masks);
+
+    py = in[0];
+    pu = in[1+(1^rgb)];
+    pv = in[1+(0^rgb)];
+
+    op = oplanes[0] + srcSliceY*outstrides[0];
+
+    for (i=0;i<h2;i++) {
+
+        lcscf (py,pu,pv,op,w,&c->oy);
+
+        py += instrides[0];
+        op += outstrides[0];
+
+        lcscf (py,pu,pv,op,w,&c->oy);
+
+        py += instrides[0];
+        pu += instrides[1];
+        pv += instrides[2];
+        op += outstrides[0];
+    }
+
+    return srcSliceH;
+}
+
+
+static int bfin_yuv420_rgb555 (SwsContext *c,
+                               uint8_t **in, int *instrides,
+                               int srcSliceY, int srcSliceH,
+                               uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb555_line, 1, 555);
+}
+
+static int bfin_yuv420_bgr555 (SwsContext *c,
+                               uint8_t **in, int *instrides,
+                               int srcSliceY, int srcSliceH,
+                               uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb555_line, 0, 555);
+}
+
+static int bfin_yuv420_rgb24 (SwsContext *c,
+                              uint8_t **in, int *instrides,
+                              int srcSliceY, int srcSliceH,
+                              uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb24_line, 1, 888);
+}
+
+static int bfin_yuv420_bgr24 (SwsContext *c,
+                              uint8_t **in, int *instrides,
+                              int srcSliceY, int srcSliceH,
+                              uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb24_line, 0, 888);
+}
+
+static int bfin_yuv420_rgb565 (SwsContext *c,
+                               uint8_t **in, int *instrides,
+                               int srcSliceY, int srcSliceH,
+                               uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb565_line, 1, 565);
+}
+
+static int bfin_yuv420_bgr565 (SwsContext *c,
+                               uint8_t **in, int *instrides,
+                               int srcSliceY, int srcSliceH,
+                               uint8_t **oplanes, int *outstrides)
+{
+    return core_yuv420_rgb (c,in,instrides,srcSliceY,srcSliceH,oplanes,outstrides,
+                            ff_bfin_yuv2rgb565_line, 0, 565);
+}
+
+
+SwsFunc ff_bfin_yuv2rgb_get_func_ptr (SwsContext *c)
+{
+    SwsFunc f;
+
+    switch(c->dstFormat) {
+    case PIX_FMT_RGB555: f = bfin_yuv420_rgb555; break;
+    case PIX_FMT_BGR555: f = bfin_yuv420_bgr555; break;
+    case PIX_FMT_RGB565: f = bfin_yuv420_rgb565; break;
+    case PIX_FMT_BGR565: f = bfin_yuv420_bgr565; break;
+    case PIX_FMT_RGB24:  f = bfin_yuv420_rgb24;  break;
+    case PIX_FMT_BGR24:  f = bfin_yuv420_bgr24;  break;
+    default:
+        return 0;
+    }
+
+    av_log(c, AV_LOG_INFO, "BlackFin Accelerated Color Space Converter %s\n",
+           sws_format_name (c->dstFormat));
+
+    return f;
+}